This invention relates to systems for inspection of substrates, especially semiconductor wafers and reticles. More specifically, the invention relates to a novel system which utilizes a sensor array in conjunction with two threshold arrangement, preferably integrated on a CMOS camera.
Several systems are known in the art for the inspection of wafers and reticles. Three examples of such systems that are currently available on the market are depicted in FIGS. 1-3. In the system exemplified in FIG. 1, the wafer 100 is illuminated with a light beam emanating from a light source 110 and reaching the wafer at a 90xc2x0 angle (generally referred to as normal illumination). Preferably, light source 110 provides coherent light, i.e., source 110 may be a laser source. The light beam is scanned over the wafer by a scanner 120, typically an acousto-optical deflector (AOD) or a rotating mirror, in the direction marked by the double-headed arrow. The wafer 100 is moved in the perpendicular direction by moving the stage upon which the wafer rests. Thus, a two dimensional area of the wafer can be scanned by the light beam.
Since the wafer has basically a mirror-like top surface, the light beam is specularly reflected back per Snell""s law at 180xc2x0. This specularly reflected light is collected by a light sensor 140 and its signal is used to obtain a xe2x80x9cbright fieldxe2x80x9d image, i.e., an image created from specularly reflected light. However, whenever the light beam hits an irregularity on the wafer, such as a particle, the light scatters in various directions. Some of the diffracted/scattered light is then collected by the light sensors 130, and their signal is used to obtain a xe2x80x9cdark fieldxe2x80x9d image, i.e., an image created from diffracted/scattered light. Thus, irregularities appear in the dark field image as stars in a dark sky.
In the system exemplified in FIG. 2, the wafer 200 is illuminated by a light beam emanating from light source 210, but reaching the wafer at a shallow angle, generally referred to as grazing illumination. The light beam is scanned over the wafer by a scanner 220, typically an acousto-optical scanner or a rotating mirror, in the direction marked by the double-headed arrow. The wafer 200 is moved in the perpendicular direction by moving the stage upon which the wafer rests. Thus, a two dimensional area of the wafer can be scanned by the light beam.
Since the light reaches the wafer at a grazing angle xcex8, its specular reflection is at a corresponding angle, xcex8xe2x80x2, according to Snell""s law. This light is collected by sensor 240, and its signal is used to create the bright field image. Any scattered light is collected by sensors 230, the signal of which is used to create dark field images.
It should be appreciated that in the above exemplified systems, with respect to each sensor the image data is acquired serially. That is, each two dimensional image, whether bright or dark field, is constructed by acquiring signals of pixel after pixel, per the scanned light beam. This is time consuming serial operation, which directly affects the throughput of such systems. Moreover, the scan speed of such systems is limited by the scanner""s speed (i.e., the band-width for an acousto-optic scanner) and by the electronics that support the detectors, e.g., the PMT (Photo-Multiplier Tube). Thus, a need exists to develop a system that does not utilize a scanned light beam.
Another difficulty with systems which use coherent light is diffraction caused by features arranged in a repeated order, thereby effectively forming a grating. Specifically, in semiconductor devices many features are constructed in a repeated order fashion. When these features are illuminated by a coherent light beam, they diffract the light in much the same manner as a diffraction grating would diffract the light. However, such constructive diffraction can be mistaken by the system for a defect. One way to overcome such a problem is to use a spatial filter in the Fourier plane, as exemplified by filters 235 in FIG. 2. This problem and proposed solutions are disclosed in, for example, U.S. Pat. Nos. 4,898,471, 4,806,774, and 5,276,498, which are incorporated herein by reference.
The system depicted in FIG. 3 performs a bright field inspection only, but does not use a scanner to scan the light beam. Instead, light source 310 provides a relatively broader light beam which illuminates the wafer 300 with a relatively large spot 315. A TDI sensor is used to image an elongated strip 325 of the illuminated spot. The length of the strip corresponds to the width of the TDI sensor. For example, if the TDI sensor comprises a 2048xc3x972048 pixels, then the scanned strip is of size 2048xc3x971. As the wafer is moved by the stage, strips are imaged and collected to form a bright field two-dimensional image of the inspected area.
Looking forward, as design rules shrink, the importance of detecting increasingly small irregularities becomes paramount. With design rules such as 0.18 and 0.15 xcexcm, very small irregularities, such as particles of sub-micron size, can be killer defects and cause the device to malfunction. However, in order to detect such small irregularities, one needs to use a very small wavelength light source, such as ultra violate (UV) or deep ultra violate (DUV) light source. This presents at least two crucial problems: first, optical elements operating in the DUV regime are expensive and, second, small short wave implies small spot size of the light beam; thus, the scanning speed and collection data rate need to be increased.
The small size of killer defects also present a formidable challenge for bright field system which do not use scanning, such as the TDI system depicted in FIG. 3. Specifically, since bright field system construct an actual image of the inspected area, the image includes multitude of structural elements built upon the wafer. Thus, the resulting image looks much like a maze, and it is increasingly difficult to detect a small irregularity in the maze-like image. Thus, the system requires a complicated image processing algorithm to recognize the defect, thereby increasing the processing power and time required and increasing the cost of purchasing and operating the system. It is not clear at this time whether even the most sophisticated algorithm may be unable to detect such small irregularities.
According to the present invention, a two dimensional sensor array is used to collect light returned from the inspected substrate. The signal generated by each individual sensor is passed through a threshold. Those signals which are below the threshold are amplified and are summed up. The summed signal is then passed through a second threshold. Summed signals which pass the second threshold are flagged as indicating suspect locations on the substrate.
The preferred implementation of the present invention is an integrated CMOS camera. Specifically, both thresholds, the amplifiers and the adder are integrated into the CMOS camera. This implementation is advantageous in that it enables much faster throughput of the system.